vectorize-io/hindsight tops today’s list. It treats agent memory as three operations: retain observations, recall them when needed, and reflect so stored knowledge stays current, with memory banks and mental models organizing what an agent knows. It deploys as Docker, pip or embedded Python against 25+ model providers, and connects to coding agents through an MCP server or a two-line wrapper. paperclipai/paperclip holds second place from a different angle: tasks arrive as tickets with approval and review gates you verify from diffs, screenshots and tests, an org chart gives every agent a role, a boss and scoped secrets, heartbeats wake agents on a schedule, and a monthly budget per agent stops runaway token spend.
If you are picking infrastructure this week, three entries earn a closer look. hydra-db/hydradb is a Rust graph database that keeps the durable graph on S3-compatible object storage, with compute split into query nodes and background indexers, and you query pinned snapshots in OpenCypher or through Neo4j drivers over Bolt. rocketride-org/rocketride-server is a visual pipeline builder for AI workloads: more than a hundred nodes cover model providers, vector databases, OCR and NER, pipelines are portable JSON drawn on a VS Code canvas, and a multithreaded C++ engine runs them on your own hardware. NandhaKishorM/laya answers typed questions about any text in one 33-millisecond forward pass, returning a choice, a score or a yes/no flag, with nothing generated and nothing to parse.
Around them, choose by situation. debpalash/VoiceStudio is a local audio workspace that puts voice design, transcription, video dubbing and long narrated projects in one Electron app, handy when you would rather try several speech workflows on your own material than chain separate tools. dream-num/univer is an embeddable Office SDK whose Facade API runs the same code in the browser and headless in Node.js, so agents can edit and verify Office files through structured APIs. latent-spaces/brag turns the project you just finished into a short launch video with one command, and deepseek-ai/deepseek-harness is DeepSeek’s plugin-based agent runtime for experimenting with how capabilities compose, keeping in mind it is still a developer preview. Chinese readers can browse eternity4719/HowToLiveBetter, a guide that prices life advice by cost, return and evidence strength.